Socio-semantic dynamics in a blog...
Transcript of Socio-semantic dynamics in a blog...
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Socio-semantic dynamics in a blog network
Jean-Philippe CointetCREA (CNRS/EP, France)
AND
Camille RothCAMS (CNRS/EHESS, France)
IEEE SOCIALCOM09, VANCOUVER, BC – AUG 29–31, 2009
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
A SOCIAL network
Three kinds of links for each blog...
citation: post citation links
interaction: comment links
affiliation: blogroll links
...where contents circulate
in terms of topics (W)
in terms of cultural items (U)
Dataset: US blogosphere
scope: 4 months of ’08 campaign
network: citations
nodes: 1, 066 blogs (RTGI)
citation link
blogroll linkcomment link
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
A socio-SEMANTIC network
Three kinds of links for each blog...
citation: post citation links
interaction: comment links
affiliation: blogroll links
...where contents circulate
in terms of topics (W)
in terms of cultural items (U)
Dataset: US blogosphere
scope: 4 months of ’08 campaign
network: citations
nodes: 1, 066 blogs (RTGI)
semantic characterization
“relevant” syntagms(“health insurance”, “climate change”, “national
security”, “super Tuesday”, “human rights”...)
urls: “www.youtube.com/x1hqwkeac”, etc.
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
A DYNAMIC socio-semantic network
Three kinds of links for each blog...
citation: post citation links
interaction: comment links
affiliation: blogroll links
...where contents circulate
in terms of topics (W)
in terms of cultural items (U)
Dataset: US blogosphere
scope: 4 months of ’08 campaign
network: citations
nodes: 1, 066 blogs (RTGI)
http://presidentialwatch08.com/
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Socio-semantic configuration
Jan 1 Michigan Primary Super Tuesday feb 170
0.01
0.02
0.03
0.04
0.05
0.06
time
freq
uen
cyof
term
s
super tuesdaymichigancaliforniahuckabee
d
c
ba
19/02
20/02
20/02
26/02
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Socio-semantic configuration
semantic profile of a blog i :
Wi (w) :=Wi (w)P|W|
w=1 Wi (w)
· log|B|
|{j, Wj (w) > 0}|
semantic distance
between blogs i and j :
δ(i , j) = 1− Wi · Wj
‖Wi‖‖Wj‖
[0;.1[ [.1;.2[ [.2;.3[ [.3;.4[ [.4;.5[ [.5;.6[ [.6;.7[ [.7;.8[ [.8;.9[ [.9;1]0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
P(δ
)
δ
Semantic distance distributions. Triangles: computed over thewhole set of possible blog pairs. Crosses: distribution computed onlinked blogs.
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Computing link creation propensity
→ estimate the “propensity of interaction”...that it is more or less likely for a node (ora dyad) with property “m” to receive a link...which may be simply estimated by:
f (m) =ν(m)
N(m)
0 50 100 150 200
100
101
k
f(k
)
ν(m) = number of links pointing towards an agent of type m(resp. number of new dyads of type m) during a time period,
N(m) = number of agents (resp. of dyads) of type m.
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Computing link creation propensity
→ estimate the “propensity of interaction”...that it is more or less likely for a node (ora dyad) with property “m” to receive a link...which may be simply estimated by:
f (m) =ν(m)
N(m)0 50 100 150 200
100
101
k
f(k
)
ν(m) = number of links pointing towards an agent of type m(resp. number of new dyads of type m) during a time period,
N(m) = number of agents (resp. of dyads) of type m.
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing
topological distance effects
→ strong trend to repetition and localinteraction
semantic distance→ strong trend to homophily
primarily “social”?
social distance and degree
0 50 100 150 200
100
101
k
f(k
)
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing
topological distance effects
→ strong trend to repetition and localinteraction
semantic distance→ strong trend to homophily
primarily “social”?
social distance and degree
1 2 3 4 >410
−2
10−1
100
101
102
d
ˆf(d
)
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing
topological distance effects
→ strong trend to repetition and localinteraction
semantic distance→ strong trend to homophily
primarily “social”?
social distance and degree
[0;.1] ].1;.2] ].2;.3] ].3;.4] ].4;.5] ].5;.6] ].6;.7] ].7;.8] ].8;.9] ].9;1]10
−1
100
101
102
δ
ˆg(δ
)
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing
topological distance effects
→ strong trend to repetition and localinteraction
semantic distance→ strong trend to homophily
primarily “social”?
social distance and degree
1 2 3 >3 [0;0.2[ [0.2;0.4[ [0.4;0.6[ [0.6;0.8[ [0.8;1]
10−4
10−3
10−2
10−1
100
semantic distance δsocial distance d
prop
ensi
onp(d
,δ)
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Dynamics of the social network
in-degree effects
→ increasing, plateauing
topological distance effects
→ strong trend to repetition and localinteraction
semantic distance→ strong trend to homophily
primarily “social”?
social distance and degree
12
3>3 0
50100
10−2
10−1
100
101
102
social capital ksocial distance d
prop
ensi
onp(d
,k)
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Information flows: measures on the post network
Dyadic measures:
raw, weighted network,aggregated on 4 months
attentional matrix a...→ and total attentionαa = 5/6
detachment matrix
“edge range”:
quantifying shortcuts d
b
c
a
1
3 2
2
13
f
e3
2
2
1
1
1
1
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Information flows: measures on the post network
Dyadic measures:
raw, weighted network,aggregated on 4 months
attentional matrix a...→ and total attentionαa = 5/6
detachment matrix
“edge range”:
quantifying shortcuts d
b
c
a
1/6
3/4 2/3
2/6
1/53/6f
e3/5
2/3
2/2
1/3
1/4
1/3
1/5
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Information flows: measures on the post network
Dyadic measures:
raw, weighted network,aggregated on 4 months
attentional matrix a...→ and total attentionαa = 5/6
detachment matrix
“edge range”:
quantifying shortcuts d
b
c
a
6
4/3 3/2
3
52f
e5/3
3/2
1
3
4
3
5
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Information flows: measures on the post network
Dyadic measures:
raw, weighted network,aggregated on 4 months
attentional matrix a...→ and total attentionαa = 5/6
detachment matrix
“edge range”:
quantifying shortcuts
d
b
c
a
6
4/3 3/2
5
2
f
e
5/3
3/2
1
3
4
3
5
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Information cascade
d
c
ba
19/02
20/02
20/02
26/02
An example of diffusion subgraph, acommon “resource” and a set ofcitation links between blogs
Diffusion subgraphs
100
101
102
100
101
102
103
104
size of diffusion subgraphs
num
ber
ofdi
ffus
ion
subg
raph
s
linksnodes
⇒ heterogeneous cascade sizes
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
An ego-centered perspective
d
c
ba
19/02
20/02
20/02
26/02
We focus on the total number of“transmissions” generated by blogswith a given total attention α
a bit more “global”...
d
c
ba
19/02
20/02
20/02
26/02
second transmissions: we focus on“later transmissions”, i.e. after a firsttransmission event
role of the total attention on thenumber of diffusion links
10−2
10−1
100
101
102
10−1
100
101
102
Total Attention αN
umber
oftr
anm
issi
ons
Larger active readership => largernumber of diffusion links, yet not linearly
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
An ego-centered perspective
d
c
ba
19/02
20/02
20/02
26/02
We focus on the total number of“transmissions” generated by blogswith a given total attention α
a bit more “global”...
d
c
ba
19/02
20/02
20/02
26/02
second transmissions: we focus on“later transmissions”, i.e. after a firsttransmission event
role of the total attention on thenumber of diffusion links
10−2
10−1
100
101
102
10−1
100
101
102
Total Attention αN
umber
oftr
anm
issi
ons
Larger active readership => largernumber of diffusion links, yet not linearly
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
A more global perspective
→ role of edge-range on thenumber of grand-children
We focus again on transmissions occurringafter a first transmission event
d
c
ba
19/02
20/02
20/02
26/02
0 10 20 30 40 50 60 70 80 90 1000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
edge range rM
ean
num
ber
of2n
dtr
ansm
issi
ons
An information which has been transmitted through a “median” linkgenerates a larger number of grandchildren
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Concluding remarks
Co-evolution of content and relationshipsPatterns not necessarily linked to authority onlyPatterns not necessarily ego-centered only→ divergent from the “neighbor-based-influence” perspective
Thank [email protected] & [email protected]
The “blogosphere” as a socio-semantic network Link creation dynamics Diffusion dynamics
Concluding remarks
Co-evolution of content and relationshipsPatterns not necessarily linked to authority onlyPatterns not necessarily ego-centered only→ divergent from the “neighbor-based-influence” perspective
Thank [email protected] & [email protected]